In geophysics, available data are generally sparse, incomplete, and ambiguous. When inferring models from these data, this results in unfavourable model properties such as high uncertainties and significant non-uniqueness. These challenges can be overcome by ensemble-generating inversion methods such as Markov chain Monte Carlo. The goal is to build a large collection of suitable models that together characterise uncertainty and non-uniqueness of the properties of interest. Though powerful, Monte Carlo methods have significant drawbacks such as long computational times and a large amount of digital outputs. The generation, handling, and distribution of model ensembles is challenging and new approaches are necessary to improve the user experience with Monte Carlo methods.
Generative models, a family of machine learning models that have attracted much attention in recent years, are one way out. They have the ability to build a parametric representation from a collection of samples, thereby enabling compression and enhancement of an ensemble. This leads to manifold advantages such as improved solution of numerical integrals which are ubiquitous in Monte Carlo inversion, and suggests possibilities to overcome the curse of dimensionality. In addition to these opportunities in dealing with existing ensembles from Monte Carlo studies, generative models offer several potential avenues during the ensemble-building process itself. These include the potential for convergence assessment during the sampling process and the possibility for building adaptive proposal distributions to make the sampling process more efficient. Many different generative models exist, and care has to be taken in the choice of a particular type, architecture, training strategies, and other hyperparameters. When overcoming these challenges, generative models promise to revolutionise the way that Monte Carlo studies in geophysical inversion are conducted in the future. This presentation explores the interaction between ensemble-based methods and generative models, and demonstrates progress towards more efficient strategies for Monte Carlo inversion.